Loading...
1#include <linux/capability.h>
2#include <linux/blkdev.h>
3#include <linux/export.h>
4#include <linux/gfp.h>
5#include <linux/blkpg.h>
6#include <linux/hdreg.h>
7#include <linux/badblocks.h>
8#include <linux/backing-dev.h>
9#include <linux/fs.h>
10#include <linux/blktrace_api.h>
11#include <linux/pr.h>
12#include <asm/uaccess.h>
13
14static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
15{
16 struct block_device *bdevp;
17 struct gendisk *disk;
18 struct hd_struct *part, *lpart;
19 struct blkpg_ioctl_arg a;
20 struct blkpg_partition p;
21 struct disk_part_iter piter;
22 long long start, length;
23 int partno;
24
25 if (!capable(CAP_SYS_ADMIN))
26 return -EACCES;
27 if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
28 return -EFAULT;
29 if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
30 return -EFAULT;
31 disk = bdev->bd_disk;
32 if (bdev != bdev->bd_contains)
33 return -EINVAL;
34 partno = p.pno;
35 if (partno <= 0)
36 return -EINVAL;
37 switch (a.op) {
38 case BLKPG_ADD_PARTITION:
39 start = p.start >> 9;
40 length = p.length >> 9;
41 /* check for fit in a hd_struct */
42 if (sizeof(sector_t) == sizeof(long) &&
43 sizeof(long long) > sizeof(long)) {
44 long pstart = start, plength = length;
45 if (pstart != start || plength != length
46 || pstart < 0 || plength < 0 || partno > 65535)
47 return -EINVAL;
48 }
49
50 mutex_lock(&bdev->bd_mutex);
51
52 /* overlap? */
53 disk_part_iter_init(&piter, disk,
54 DISK_PITER_INCL_EMPTY);
55 while ((part = disk_part_iter_next(&piter))) {
56 if (!(start + length <= part->start_sect ||
57 start >= part->start_sect + part->nr_sects)) {
58 disk_part_iter_exit(&piter);
59 mutex_unlock(&bdev->bd_mutex);
60 return -EBUSY;
61 }
62 }
63 disk_part_iter_exit(&piter);
64
65 /* all seems OK */
66 part = add_partition(disk, partno, start, length,
67 ADDPART_FLAG_NONE, NULL);
68 mutex_unlock(&bdev->bd_mutex);
69 return PTR_ERR_OR_ZERO(part);
70 case BLKPG_DEL_PARTITION:
71 part = disk_get_part(disk, partno);
72 if (!part)
73 return -ENXIO;
74
75 bdevp = bdget(part_devt(part));
76 disk_put_part(part);
77 if (!bdevp)
78 return -ENOMEM;
79
80 mutex_lock(&bdevp->bd_mutex);
81 if (bdevp->bd_openers) {
82 mutex_unlock(&bdevp->bd_mutex);
83 bdput(bdevp);
84 return -EBUSY;
85 }
86 /* all seems OK */
87 fsync_bdev(bdevp);
88 invalidate_bdev(bdevp);
89
90 mutex_lock_nested(&bdev->bd_mutex, 1);
91 delete_partition(disk, partno);
92 mutex_unlock(&bdev->bd_mutex);
93 mutex_unlock(&bdevp->bd_mutex);
94 bdput(bdevp);
95
96 return 0;
97 case BLKPG_RESIZE_PARTITION:
98 start = p.start >> 9;
99 /* new length of partition in bytes */
100 length = p.length >> 9;
101 /* check for fit in a hd_struct */
102 if (sizeof(sector_t) == sizeof(long) &&
103 sizeof(long long) > sizeof(long)) {
104 long pstart = start, plength = length;
105 if (pstart != start || plength != length
106 || pstart < 0 || plength < 0)
107 return -EINVAL;
108 }
109 part = disk_get_part(disk, partno);
110 if (!part)
111 return -ENXIO;
112 bdevp = bdget(part_devt(part));
113 if (!bdevp) {
114 disk_put_part(part);
115 return -ENOMEM;
116 }
117 mutex_lock(&bdevp->bd_mutex);
118 mutex_lock_nested(&bdev->bd_mutex, 1);
119 if (start != part->start_sect) {
120 mutex_unlock(&bdevp->bd_mutex);
121 mutex_unlock(&bdev->bd_mutex);
122 bdput(bdevp);
123 disk_put_part(part);
124 return -EINVAL;
125 }
126 /* overlap? */
127 disk_part_iter_init(&piter, disk,
128 DISK_PITER_INCL_EMPTY);
129 while ((lpart = disk_part_iter_next(&piter))) {
130 if (lpart->partno != partno &&
131 !(start + length <= lpart->start_sect ||
132 start >= lpart->start_sect + lpart->nr_sects)
133 ) {
134 disk_part_iter_exit(&piter);
135 mutex_unlock(&bdevp->bd_mutex);
136 mutex_unlock(&bdev->bd_mutex);
137 bdput(bdevp);
138 disk_put_part(part);
139 return -EBUSY;
140 }
141 }
142 disk_part_iter_exit(&piter);
143 part_nr_sects_write(part, (sector_t)length);
144 i_size_write(bdevp->bd_inode, p.length);
145 mutex_unlock(&bdevp->bd_mutex);
146 mutex_unlock(&bdev->bd_mutex);
147 bdput(bdevp);
148 disk_put_part(part);
149 return 0;
150 default:
151 return -EINVAL;
152 }
153}
154
155/*
156 * This is an exported API for the block driver, and will not
157 * acquire bd_mutex. This API should be used in case that
158 * caller has held bd_mutex already.
159 */
160int __blkdev_reread_part(struct block_device *bdev)
161{
162 struct gendisk *disk = bdev->bd_disk;
163
164 if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
165 return -EINVAL;
166 if (!capable(CAP_SYS_ADMIN))
167 return -EACCES;
168
169 lockdep_assert_held(&bdev->bd_mutex);
170
171 return rescan_partitions(disk, bdev);
172}
173EXPORT_SYMBOL(__blkdev_reread_part);
174
175/*
176 * This is an exported API for the block driver, and will
177 * try to acquire bd_mutex. If bd_mutex has been held already
178 * in current context, please call __blkdev_reread_part().
179 *
180 * Make sure the held locks in current context aren't required
181 * in open()/close() handler and I/O path for avoiding ABBA deadlock:
182 * - bd_mutex is held before calling block driver's open/close
183 * handler
184 * - reading partition table may submit I/O to the block device
185 */
186int blkdev_reread_part(struct block_device *bdev)
187{
188 int res;
189
190 mutex_lock(&bdev->bd_mutex);
191 res = __blkdev_reread_part(bdev);
192 mutex_unlock(&bdev->bd_mutex);
193
194 return res;
195}
196EXPORT_SYMBOL(blkdev_reread_part);
197
198static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
199 unsigned long arg, unsigned long flags)
200{
201 uint64_t range[2];
202 uint64_t start, len;
203
204 if (!(mode & FMODE_WRITE))
205 return -EBADF;
206
207 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
208 return -EFAULT;
209
210 start = range[0];
211 len = range[1];
212
213 if (start & 511)
214 return -EINVAL;
215 if (len & 511)
216 return -EINVAL;
217 start >>= 9;
218 len >>= 9;
219
220 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
221 return -EINVAL;
222 return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
223}
224
225static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
226 unsigned long arg)
227{
228 uint64_t range[2];
229 uint64_t start, len;
230
231 if (!(mode & FMODE_WRITE))
232 return -EBADF;
233
234 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
235 return -EFAULT;
236
237 start = range[0];
238 len = range[1];
239
240 if (start & 511)
241 return -EINVAL;
242 if (len & 511)
243 return -EINVAL;
244 start >>= 9;
245 len >>= 9;
246
247 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
248 return -EINVAL;
249
250 return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL, false);
251}
252
253static int put_ushort(unsigned long arg, unsigned short val)
254{
255 return put_user(val, (unsigned short __user *)arg);
256}
257
258static int put_int(unsigned long arg, int val)
259{
260 return put_user(val, (int __user *)arg);
261}
262
263static int put_uint(unsigned long arg, unsigned int val)
264{
265 return put_user(val, (unsigned int __user *)arg);
266}
267
268static int put_long(unsigned long arg, long val)
269{
270 return put_user(val, (long __user *)arg);
271}
272
273static int put_ulong(unsigned long arg, unsigned long val)
274{
275 return put_user(val, (unsigned long __user *)arg);
276}
277
278static int put_u64(unsigned long arg, u64 val)
279{
280 return put_user(val, (u64 __user *)arg);
281}
282
283int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
284 unsigned cmd, unsigned long arg)
285{
286 struct gendisk *disk = bdev->bd_disk;
287
288 if (disk->fops->ioctl)
289 return disk->fops->ioctl(bdev, mode, cmd, arg);
290
291 return -ENOTTY;
292}
293/*
294 * For the record: _GPL here is only because somebody decided to slap it
295 * on the previous export. Sheer idiocy, since it wasn't copyrightable
296 * at all and could be open-coded without any exports by anybody who cares.
297 */
298EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
299
300static int blkdev_pr_register(struct block_device *bdev,
301 struct pr_registration __user *arg)
302{
303 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
304 struct pr_registration reg;
305
306 if (!capable(CAP_SYS_ADMIN))
307 return -EPERM;
308 if (!ops || !ops->pr_register)
309 return -EOPNOTSUPP;
310 if (copy_from_user(®, arg, sizeof(reg)))
311 return -EFAULT;
312
313 if (reg.flags & ~PR_FL_IGNORE_KEY)
314 return -EOPNOTSUPP;
315 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
316}
317
318static int blkdev_pr_reserve(struct block_device *bdev,
319 struct pr_reservation __user *arg)
320{
321 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
322 struct pr_reservation rsv;
323
324 if (!capable(CAP_SYS_ADMIN))
325 return -EPERM;
326 if (!ops || !ops->pr_reserve)
327 return -EOPNOTSUPP;
328 if (copy_from_user(&rsv, arg, sizeof(rsv)))
329 return -EFAULT;
330
331 if (rsv.flags & ~PR_FL_IGNORE_KEY)
332 return -EOPNOTSUPP;
333 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
334}
335
336static int blkdev_pr_release(struct block_device *bdev,
337 struct pr_reservation __user *arg)
338{
339 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
340 struct pr_reservation rsv;
341
342 if (!capable(CAP_SYS_ADMIN))
343 return -EPERM;
344 if (!ops || !ops->pr_release)
345 return -EOPNOTSUPP;
346 if (copy_from_user(&rsv, arg, sizeof(rsv)))
347 return -EFAULT;
348
349 if (rsv.flags)
350 return -EOPNOTSUPP;
351 return ops->pr_release(bdev, rsv.key, rsv.type);
352}
353
354static int blkdev_pr_preempt(struct block_device *bdev,
355 struct pr_preempt __user *arg, bool abort)
356{
357 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
358 struct pr_preempt p;
359
360 if (!capable(CAP_SYS_ADMIN))
361 return -EPERM;
362 if (!ops || !ops->pr_preempt)
363 return -EOPNOTSUPP;
364 if (copy_from_user(&p, arg, sizeof(p)))
365 return -EFAULT;
366
367 if (p.flags)
368 return -EOPNOTSUPP;
369 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
370}
371
372static int blkdev_pr_clear(struct block_device *bdev,
373 struct pr_clear __user *arg)
374{
375 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
376 struct pr_clear c;
377
378 if (!capable(CAP_SYS_ADMIN))
379 return -EPERM;
380 if (!ops || !ops->pr_clear)
381 return -EOPNOTSUPP;
382 if (copy_from_user(&c, arg, sizeof(c)))
383 return -EFAULT;
384
385 if (c.flags)
386 return -EOPNOTSUPP;
387 return ops->pr_clear(bdev, c.key);
388}
389
390/*
391 * Is it an unrecognized ioctl? The correct returns are either
392 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
393 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
394 * code before returning.
395 *
396 * Confused drivers sometimes return EINVAL, which is wrong. It
397 * means "I understood the ioctl command, but the parameters to
398 * it were wrong".
399 *
400 * We should aim to just fix the broken drivers, the EINVAL case
401 * should go away.
402 */
403static inline int is_unrecognized_ioctl(int ret)
404{
405 return ret == -EINVAL ||
406 ret == -ENOTTY ||
407 ret == -ENOIOCTLCMD;
408}
409
410#ifdef CONFIG_FS_DAX
411bool blkdev_dax_capable(struct block_device *bdev)
412{
413 struct gendisk *disk = bdev->bd_disk;
414
415 if (!disk->fops->direct_access)
416 return false;
417
418 /*
419 * If the partition is not aligned on a page boundary, we can't
420 * do dax I/O to it.
421 */
422 if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512))
423 || (bdev->bd_part->nr_sects % (PAGE_SIZE / 512)))
424 return false;
425
426 /*
427 * If the device has known bad blocks, force all I/O through the
428 * driver / page cache.
429 *
430 * TODO: support finer grained dax error handling
431 */
432 if (disk->bb && disk->bb->count)
433 return false;
434
435 return true;
436}
437#endif
438
439static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
440 unsigned cmd, unsigned long arg)
441{
442 int ret;
443
444 if (!capable(CAP_SYS_ADMIN))
445 return -EACCES;
446
447 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
448 if (!is_unrecognized_ioctl(ret))
449 return ret;
450
451 fsync_bdev(bdev);
452 invalidate_bdev(bdev);
453 return 0;
454}
455
456static int blkdev_roset(struct block_device *bdev, fmode_t mode,
457 unsigned cmd, unsigned long arg)
458{
459 int ret, n;
460
461 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
462 if (!is_unrecognized_ioctl(ret))
463 return ret;
464 if (!capable(CAP_SYS_ADMIN))
465 return -EACCES;
466 if (get_user(n, (int __user *)arg))
467 return -EFAULT;
468 set_device_ro(bdev, n);
469 return 0;
470}
471
472static int blkdev_getgeo(struct block_device *bdev,
473 struct hd_geometry __user *argp)
474{
475 struct gendisk *disk = bdev->bd_disk;
476 struct hd_geometry geo;
477 int ret;
478
479 if (!argp)
480 return -EINVAL;
481 if (!disk->fops->getgeo)
482 return -ENOTTY;
483
484 /*
485 * We need to set the startsect first, the driver may
486 * want to override it.
487 */
488 memset(&geo, 0, sizeof(geo));
489 geo.start = get_start_sect(bdev);
490 ret = disk->fops->getgeo(bdev, &geo);
491 if (ret)
492 return ret;
493 if (copy_to_user(argp, &geo, sizeof(geo)))
494 return -EFAULT;
495 return 0;
496}
497
498/* set the logical block size */
499static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
500 int __user *argp)
501{
502 int ret, n;
503
504 if (!capable(CAP_SYS_ADMIN))
505 return -EACCES;
506 if (!argp)
507 return -EINVAL;
508 if (get_user(n, argp))
509 return -EFAULT;
510
511 if (!(mode & FMODE_EXCL)) {
512 bdgrab(bdev);
513 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
514 return -EBUSY;
515 }
516
517 ret = set_blocksize(bdev, n);
518 if (!(mode & FMODE_EXCL))
519 blkdev_put(bdev, mode | FMODE_EXCL);
520 return ret;
521}
522
523/*
524 * always keep this in sync with compat_blkdev_ioctl()
525 */
526int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
527 unsigned long arg)
528{
529 struct backing_dev_info *bdi;
530 void __user *argp = (void __user *)arg;
531 loff_t size;
532 unsigned int max_sectors;
533
534 switch (cmd) {
535 case BLKFLSBUF:
536 return blkdev_flushbuf(bdev, mode, cmd, arg);
537 case BLKROSET:
538 return blkdev_roset(bdev, mode, cmd, arg);
539 case BLKDISCARD:
540 return blk_ioctl_discard(bdev, mode, arg, 0);
541 case BLKSECDISCARD:
542 return blk_ioctl_discard(bdev, mode, arg,
543 BLKDEV_DISCARD_SECURE);
544 case BLKZEROOUT:
545 return blk_ioctl_zeroout(bdev, mode, arg);
546 case HDIO_GETGEO:
547 return blkdev_getgeo(bdev, argp);
548 case BLKRAGET:
549 case BLKFRAGET:
550 if (!arg)
551 return -EINVAL;
552 bdi = blk_get_backing_dev_info(bdev);
553 return put_long(arg, (bdi->ra_pages * PAGE_SIZE) / 512);
554 case BLKROGET:
555 return put_int(arg, bdev_read_only(bdev) != 0);
556 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
557 return put_int(arg, block_size(bdev));
558 case BLKSSZGET: /* get block device logical block size */
559 return put_int(arg, bdev_logical_block_size(bdev));
560 case BLKPBSZGET: /* get block device physical block size */
561 return put_uint(arg, bdev_physical_block_size(bdev));
562 case BLKIOMIN:
563 return put_uint(arg, bdev_io_min(bdev));
564 case BLKIOOPT:
565 return put_uint(arg, bdev_io_opt(bdev));
566 case BLKALIGNOFF:
567 return put_int(arg, bdev_alignment_offset(bdev));
568 case BLKDISCARDZEROES:
569 return put_uint(arg, bdev_discard_zeroes_data(bdev));
570 case BLKSECTGET:
571 max_sectors = min_t(unsigned int, USHRT_MAX,
572 queue_max_sectors(bdev_get_queue(bdev)));
573 return put_ushort(arg, max_sectors);
574 case BLKROTATIONAL:
575 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
576 case BLKRASET:
577 case BLKFRASET:
578 if(!capable(CAP_SYS_ADMIN))
579 return -EACCES;
580 bdi = blk_get_backing_dev_info(bdev);
581 bdi->ra_pages = (arg * 512) / PAGE_SIZE;
582 return 0;
583 case BLKBSZSET:
584 return blkdev_bszset(bdev, mode, argp);
585 case BLKPG:
586 return blkpg_ioctl(bdev, argp);
587 case BLKRRPART:
588 return blkdev_reread_part(bdev);
589 case BLKGETSIZE:
590 size = i_size_read(bdev->bd_inode);
591 if ((size >> 9) > ~0UL)
592 return -EFBIG;
593 return put_ulong(arg, size >> 9);
594 case BLKGETSIZE64:
595 return put_u64(arg, i_size_read(bdev->bd_inode));
596 case BLKTRACESTART:
597 case BLKTRACESTOP:
598 case BLKTRACESETUP:
599 case BLKTRACETEARDOWN:
600 return blk_trace_ioctl(bdev, cmd, argp);
601 case BLKDAXGET:
602 return put_int(arg, !!(bdev->bd_inode->i_flags & S_DAX));
603 break;
604 case IOC_PR_REGISTER:
605 return blkdev_pr_register(bdev, argp);
606 case IOC_PR_RESERVE:
607 return blkdev_pr_reserve(bdev, argp);
608 case IOC_PR_RELEASE:
609 return blkdev_pr_release(bdev, argp);
610 case IOC_PR_PREEMPT:
611 return blkdev_pr_preempt(bdev, argp, false);
612 case IOC_PR_PREEMPT_ABORT:
613 return blkdev_pr_preempt(bdev, argp, true);
614 case IOC_PR_CLEAR:
615 return blkdev_pr_clear(bdev, argp);
616 default:
617 return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
618 }
619}
620EXPORT_SYMBOL_GPL(blkdev_ioctl);
1#include <linux/capability.h>
2#include <linux/blkdev.h>
3#include <linux/export.h>
4#include <linux/gfp.h>
5#include <linux/blkpg.h>
6#include <linux/hdreg.h>
7#include <linux/backing-dev.h>
8#include <linux/fs.h>
9#include <linux/blktrace_api.h>
10#include <asm/uaccess.h>
11
12static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
13{
14 struct block_device *bdevp;
15 struct gendisk *disk;
16 struct hd_struct *part, *lpart;
17 struct blkpg_ioctl_arg a;
18 struct blkpg_partition p;
19 struct disk_part_iter piter;
20 long long start, length;
21 int partno;
22
23 if (!capable(CAP_SYS_ADMIN))
24 return -EACCES;
25 if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
26 return -EFAULT;
27 if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
28 return -EFAULT;
29 disk = bdev->bd_disk;
30 if (bdev != bdev->bd_contains)
31 return -EINVAL;
32 partno = p.pno;
33 if (partno <= 0)
34 return -EINVAL;
35 switch (a.op) {
36 case BLKPG_ADD_PARTITION:
37 start = p.start >> 9;
38 length = p.length >> 9;
39 /* check for fit in a hd_struct */
40 if (sizeof(sector_t) == sizeof(long) &&
41 sizeof(long long) > sizeof(long)) {
42 long pstart = start, plength = length;
43 if (pstart != start || plength != length
44 || pstart < 0 || plength < 0 || partno > 65535)
45 return -EINVAL;
46 }
47
48 mutex_lock(&bdev->bd_mutex);
49
50 /* overlap? */
51 disk_part_iter_init(&piter, disk,
52 DISK_PITER_INCL_EMPTY);
53 while ((part = disk_part_iter_next(&piter))) {
54 if (!(start + length <= part->start_sect ||
55 start >= part->start_sect + part->nr_sects)) {
56 disk_part_iter_exit(&piter);
57 mutex_unlock(&bdev->bd_mutex);
58 return -EBUSY;
59 }
60 }
61 disk_part_iter_exit(&piter);
62
63 /* all seems OK */
64 part = add_partition(disk, partno, start, length,
65 ADDPART_FLAG_NONE, NULL);
66 mutex_unlock(&bdev->bd_mutex);
67 return PTR_ERR_OR_ZERO(part);
68 case BLKPG_DEL_PARTITION:
69 part = disk_get_part(disk, partno);
70 if (!part)
71 return -ENXIO;
72
73 bdevp = bdget(part_devt(part));
74 disk_put_part(part);
75 if (!bdevp)
76 return -ENOMEM;
77
78 mutex_lock(&bdevp->bd_mutex);
79 if (bdevp->bd_openers) {
80 mutex_unlock(&bdevp->bd_mutex);
81 bdput(bdevp);
82 return -EBUSY;
83 }
84 /* all seems OK */
85 fsync_bdev(bdevp);
86 invalidate_bdev(bdevp);
87
88 mutex_lock_nested(&bdev->bd_mutex, 1);
89 delete_partition(disk, partno);
90 mutex_unlock(&bdev->bd_mutex);
91 mutex_unlock(&bdevp->bd_mutex);
92 bdput(bdevp);
93
94 return 0;
95 case BLKPG_RESIZE_PARTITION:
96 start = p.start >> 9;
97 /* new length of partition in bytes */
98 length = p.length >> 9;
99 /* check for fit in a hd_struct */
100 if (sizeof(sector_t) == sizeof(long) &&
101 sizeof(long long) > sizeof(long)) {
102 long pstart = start, plength = length;
103 if (pstart != start || plength != length
104 || pstart < 0 || plength < 0)
105 return -EINVAL;
106 }
107 part = disk_get_part(disk, partno);
108 if (!part)
109 return -ENXIO;
110 bdevp = bdget(part_devt(part));
111 if (!bdevp) {
112 disk_put_part(part);
113 return -ENOMEM;
114 }
115 mutex_lock(&bdevp->bd_mutex);
116 mutex_lock_nested(&bdev->bd_mutex, 1);
117 if (start != part->start_sect) {
118 mutex_unlock(&bdevp->bd_mutex);
119 mutex_unlock(&bdev->bd_mutex);
120 bdput(bdevp);
121 disk_put_part(part);
122 return -EINVAL;
123 }
124 /* overlap? */
125 disk_part_iter_init(&piter, disk,
126 DISK_PITER_INCL_EMPTY);
127 while ((lpart = disk_part_iter_next(&piter))) {
128 if (lpart->partno != partno &&
129 !(start + length <= lpart->start_sect ||
130 start >= lpart->start_sect + lpart->nr_sects)
131 ) {
132 disk_part_iter_exit(&piter);
133 mutex_unlock(&bdevp->bd_mutex);
134 mutex_unlock(&bdev->bd_mutex);
135 bdput(bdevp);
136 disk_put_part(part);
137 return -EBUSY;
138 }
139 }
140 disk_part_iter_exit(&piter);
141 part_nr_sects_write(part, (sector_t)length);
142 i_size_write(bdevp->bd_inode, p.length);
143 mutex_unlock(&bdevp->bd_mutex);
144 mutex_unlock(&bdev->bd_mutex);
145 bdput(bdevp);
146 disk_put_part(part);
147 return 0;
148 default:
149 return -EINVAL;
150 }
151}
152
153static int blkdev_reread_part(struct block_device *bdev)
154{
155 struct gendisk *disk = bdev->bd_disk;
156 int res;
157
158 if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
159 return -EINVAL;
160 if (!capable(CAP_SYS_ADMIN))
161 return -EACCES;
162 if (!mutex_trylock(&bdev->bd_mutex))
163 return -EBUSY;
164 res = rescan_partitions(disk, bdev);
165 mutex_unlock(&bdev->bd_mutex);
166 return res;
167}
168
169static int blk_ioctl_discard(struct block_device *bdev, uint64_t start,
170 uint64_t len, int secure)
171{
172 unsigned long flags = 0;
173
174 if (start & 511)
175 return -EINVAL;
176 if (len & 511)
177 return -EINVAL;
178 start >>= 9;
179 len >>= 9;
180
181 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
182 return -EINVAL;
183 if (secure)
184 flags |= BLKDEV_DISCARD_SECURE;
185 return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
186}
187
188static int blk_ioctl_zeroout(struct block_device *bdev, uint64_t start,
189 uint64_t len)
190{
191 if (start & 511)
192 return -EINVAL;
193 if (len & 511)
194 return -EINVAL;
195 start >>= 9;
196 len >>= 9;
197
198 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
199 return -EINVAL;
200
201 return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL);
202}
203
204static int put_ushort(unsigned long arg, unsigned short val)
205{
206 return put_user(val, (unsigned short __user *)arg);
207}
208
209static int put_int(unsigned long arg, int val)
210{
211 return put_user(val, (int __user *)arg);
212}
213
214static int put_uint(unsigned long arg, unsigned int val)
215{
216 return put_user(val, (unsigned int __user *)arg);
217}
218
219static int put_long(unsigned long arg, long val)
220{
221 return put_user(val, (long __user *)arg);
222}
223
224static int put_ulong(unsigned long arg, unsigned long val)
225{
226 return put_user(val, (unsigned long __user *)arg);
227}
228
229static int put_u64(unsigned long arg, u64 val)
230{
231 return put_user(val, (u64 __user *)arg);
232}
233
234int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
235 unsigned cmd, unsigned long arg)
236{
237 struct gendisk *disk = bdev->bd_disk;
238
239 if (disk->fops->ioctl)
240 return disk->fops->ioctl(bdev, mode, cmd, arg);
241
242 return -ENOTTY;
243}
244/*
245 * For the record: _GPL here is only because somebody decided to slap it
246 * on the previous export. Sheer idiocy, since it wasn't copyrightable
247 * at all and could be open-coded without any exports by anybody who cares.
248 */
249EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
250
251/*
252 * Is it an unrecognized ioctl? The correct returns are either
253 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
254 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
255 * code before returning.
256 *
257 * Confused drivers sometimes return EINVAL, which is wrong. It
258 * means "I understood the ioctl command, but the parameters to
259 * it were wrong".
260 *
261 * We should aim to just fix the broken drivers, the EINVAL case
262 * should go away.
263 */
264static inline int is_unrecognized_ioctl(int ret)
265{
266 return ret == -EINVAL ||
267 ret == -ENOTTY ||
268 ret == -ENOIOCTLCMD;
269}
270
271/*
272 * always keep this in sync with compat_blkdev_ioctl()
273 */
274int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
275 unsigned long arg)
276{
277 struct gendisk *disk = bdev->bd_disk;
278 struct backing_dev_info *bdi;
279 loff_t size;
280 int ret, n;
281
282 switch(cmd) {
283 case BLKFLSBUF:
284 if (!capable(CAP_SYS_ADMIN))
285 return -EACCES;
286
287 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
288 if (!is_unrecognized_ioctl(ret))
289 return ret;
290
291 fsync_bdev(bdev);
292 invalidate_bdev(bdev);
293 return 0;
294
295 case BLKROSET:
296 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
297 if (!is_unrecognized_ioctl(ret))
298 return ret;
299 if (!capable(CAP_SYS_ADMIN))
300 return -EACCES;
301 if (get_user(n, (int __user *)(arg)))
302 return -EFAULT;
303 set_device_ro(bdev, n);
304 return 0;
305
306 case BLKDISCARD:
307 case BLKSECDISCARD: {
308 uint64_t range[2];
309
310 if (!(mode & FMODE_WRITE))
311 return -EBADF;
312
313 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
314 return -EFAULT;
315
316 return blk_ioctl_discard(bdev, range[0], range[1],
317 cmd == BLKSECDISCARD);
318 }
319 case BLKZEROOUT: {
320 uint64_t range[2];
321
322 if (!(mode & FMODE_WRITE))
323 return -EBADF;
324
325 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
326 return -EFAULT;
327
328 return blk_ioctl_zeroout(bdev, range[0], range[1]);
329 }
330
331 case HDIO_GETGEO: {
332 struct hd_geometry geo;
333
334 if (!arg)
335 return -EINVAL;
336 if (!disk->fops->getgeo)
337 return -ENOTTY;
338
339 /*
340 * We need to set the startsect first, the driver may
341 * want to override it.
342 */
343 memset(&geo, 0, sizeof(geo));
344 geo.start = get_start_sect(bdev);
345 ret = disk->fops->getgeo(bdev, &geo);
346 if (ret)
347 return ret;
348 if (copy_to_user((struct hd_geometry __user *)arg, &geo,
349 sizeof(geo)))
350 return -EFAULT;
351 return 0;
352 }
353 case BLKRAGET:
354 case BLKFRAGET:
355 if (!arg)
356 return -EINVAL;
357 bdi = blk_get_backing_dev_info(bdev);
358 if (bdi == NULL)
359 return -ENOTTY;
360 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
361 case BLKROGET:
362 return put_int(arg, bdev_read_only(bdev) != 0);
363 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
364 return put_int(arg, block_size(bdev));
365 case BLKSSZGET: /* get block device logical block size */
366 return put_int(arg, bdev_logical_block_size(bdev));
367 case BLKPBSZGET: /* get block device physical block size */
368 return put_uint(arg, bdev_physical_block_size(bdev));
369 case BLKIOMIN:
370 return put_uint(arg, bdev_io_min(bdev));
371 case BLKIOOPT:
372 return put_uint(arg, bdev_io_opt(bdev));
373 case BLKALIGNOFF:
374 return put_int(arg, bdev_alignment_offset(bdev));
375 case BLKDISCARDZEROES:
376 return put_uint(arg, bdev_discard_zeroes_data(bdev));
377 case BLKSECTGET:
378 return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
379 case BLKROTATIONAL:
380 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
381 case BLKRASET:
382 case BLKFRASET:
383 if(!capable(CAP_SYS_ADMIN))
384 return -EACCES;
385 bdi = blk_get_backing_dev_info(bdev);
386 if (bdi == NULL)
387 return -ENOTTY;
388 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
389 return 0;
390 case BLKBSZSET:
391 /* set the logical block size */
392 if (!capable(CAP_SYS_ADMIN))
393 return -EACCES;
394 if (!arg)
395 return -EINVAL;
396 if (get_user(n, (int __user *) arg))
397 return -EFAULT;
398 if (!(mode & FMODE_EXCL)) {
399 bdgrab(bdev);
400 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
401 return -EBUSY;
402 }
403 ret = set_blocksize(bdev, n);
404 if (!(mode & FMODE_EXCL))
405 blkdev_put(bdev, mode | FMODE_EXCL);
406 return ret;
407 case BLKPG:
408 ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
409 break;
410 case BLKRRPART:
411 ret = blkdev_reread_part(bdev);
412 break;
413 case BLKGETSIZE:
414 size = i_size_read(bdev->bd_inode);
415 if ((size >> 9) > ~0UL)
416 return -EFBIG;
417 return put_ulong(arg, size >> 9);
418 case BLKGETSIZE64:
419 return put_u64(arg, i_size_read(bdev->bd_inode));
420 case BLKTRACESTART:
421 case BLKTRACESTOP:
422 case BLKTRACESETUP:
423 case BLKTRACETEARDOWN:
424 ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
425 break;
426 default:
427 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
428 }
429 return ret;
430}
431EXPORT_SYMBOL_GPL(blkdev_ioctl);